4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd-zfs/osd_object.c
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mike Pershin <tappro@whamcloud.com>
36 * Author: Johann Lombardi <johann@whamcloud.com>
39 #define DEBUG_SUBSYSTEM S_OSD
41 #include <lustre_ver.h>
42 #include <libcfs/libcfs.h>
43 #include <obd_support.h>
44 #include <lustre_net.h>
46 #include <obd_class.h>
47 #include <lustre_disk.h>
48 #include <lustre_fid.h>
50 #include "osd_internal.h"
52 #include <sys/dnode.h>
57 #include <sys/spa_impl.h>
58 #include <sys/zfs_znode.h>
59 #include <sys/dmu_tx.h>
60 #include <sys/dmu_objset.h>
61 #include <sys/dsl_prop.h>
62 #include <sys/sa_impl.h>
65 char *osd_obj_tag = "osd_object";
67 static struct dt_object_operations osd_obj_ops;
68 static struct lu_object_operations osd_lu_obj_ops;
69 extern struct dt_body_operations osd_body_ops;
70 static struct dt_object_operations osd_obj_otable_it_ops;
72 extern struct kmem_cache *osd_object_kmem;
75 osd_object_sa_fini(struct osd_object *obj)
78 sa_handle_destroy(obj->oo_sa_hdl);
79 obj->oo_sa_hdl = NULL;
84 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
88 LASSERT(obj->oo_sa_hdl == NULL);
89 LASSERT(obj->oo_dn != NULL);
91 rc = osd_sa_handle_get(obj);
95 /* Cache the xattr object id, valid for the life of the object */
96 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
98 obj->oo_xattr = ZFS_NO_OBJECT;
101 osd_object_sa_fini(obj);
108 * Add object to list of dirty objects in tx handle.
111 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
113 if (!list_empty(&obj->oo_sa_linkage))
116 down(&oh->ot_sa_lock);
117 write_lock(&obj->oo_attr_lock);
118 if (likely(list_empty(&obj->oo_sa_linkage)))
119 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
120 write_unlock(&obj->oo_attr_lock);
125 * Release spill block dbuf hold for all dirty SAs.
127 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
129 struct osd_object *obj;
131 down(&oh->ot_sa_lock);
132 while (!list_empty(&oh->ot_sa_list)) {
133 obj = list_entry(oh->ot_sa_list.next,
134 struct osd_object, oo_sa_linkage);
135 sa_spill_rele(obj->oo_sa_hdl);
136 write_lock(&obj->oo_attr_lock);
137 list_del_init(&obj->oo_sa_linkage);
138 write_unlock(&obj->oo_attr_lock);
144 * Update the SA and add the object to the dirty list.
146 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
147 void *buf, uint32_t buflen, struct osd_thandle *oh)
151 LASSERT(obj->oo_sa_hdl != NULL);
152 LASSERT(oh->ot_tx != NULL);
154 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
155 osd_object_sa_dirty_add(obj, oh);
161 * Bulk update the SA and add the object to the dirty list.
164 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
165 int count, struct osd_thandle *oh)
169 LASSERT(obj->oo_sa_hdl != NULL);
170 LASSERT(oh->ot_tx != NULL);
172 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
173 osd_object_sa_dirty_add(obj, oh);
179 * Retrieve the attributes of a DMU object
181 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
182 struct osd_object *obj, struct lu_attr *la)
184 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
185 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
190 LASSERT(obj->oo_dn != NULL);
192 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
193 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
195 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
196 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
197 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
198 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
204 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
206 rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
210 la->la_atime = osa->atime[0];
211 la->la_mtime = osa->mtime[0];
212 la->la_ctime = osa->ctime[0];
213 la->la_mode = osa->mode;
214 la->la_uid = osa->uid;
215 la->la_gid = osa->gid;
216 la->la_nlink = osa->nlink;
217 la->la_flags = attrs_zfs2fs(osa->flags);
218 la->la_size = osa->size;
220 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
221 * flags is stored in LMA, and it is only for orphan directory */
222 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
223 struct osd_thread_info *info = osd_oti_get(env);
224 struct lustre_mdt_attrs *lma;
227 lma = (struct lustre_mdt_attrs *)info->oti_buf;
229 buf.lb_len = sizeof(info->oti_buf);
230 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
233 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
235 lma_to_lustre_flags(lma->lma_incompat);
237 } else if (rc == -ENODATA) {
242 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
243 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
246 la->la_rdev = osa->rdev;
247 la->la_valid |= LA_RDEV;
254 int __osd_obj2dnode(const struct lu_env *env, objset_t *os,
255 uint64_t oid, dnode_t **dnp)
257 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
262 rc = dmu_bonus_hold(os, oid, osd_obj_tag, &db);
266 dbi = (dmu_buf_impl_t *)db;
268 *dnp = DB_DNODE(dbi);
270 LASSERT(*dnp != NULL);
271 dmu_object_info_from_dnode(*dnp, doi);
272 if (unlikely (oid != DMU_USERUSED_OBJECT &&
273 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
274 osd_dnode_rele(*dnp);
283 * Concurrency: no concurrent access is possible that early in object
286 struct lu_object *osd_object_alloc(const struct lu_env *env,
287 const struct lu_object_header *hdr,
290 struct osd_object *mo;
292 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
296 l = &mo->oo_dt.do_lu;
297 dt_object_init(&mo->oo_dt, NULL, d);
298 mo->oo_dt.do_ops = &osd_obj_ops;
299 l->lo_ops = &osd_lu_obj_ops;
300 INIT_LIST_HEAD(&mo->oo_sa_linkage);
301 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
302 init_rwsem(&mo->oo_sem);
303 init_rwsem(&mo->oo_guard);
304 rwlock_init(&mo->oo_attr_lock);
305 mo->oo_destroy = OSD_DESTROY_NONE;
313 * Concurrency: shouldn't matter.
315 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
317 struct osd_device *osd = osd_obj2dev(obj);
318 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
322 if (obj->oo_dn == NULL)
327 rc = osd_object_sa_init(obj, osd);
331 /* cache attrs in object */
332 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
336 if (likely(!fid_is_acct(fid)))
337 /* no body operations for accounting objects */
338 obj->oo_dt.do_body_ops = &osd_body_ops;
341 * initialize object before marking it existing
343 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
346 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
351 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
353 struct osd_thread_info *info = osd_oti_get(env);
356 struct lustre_mdt_attrs *lma;
359 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
360 lma = (struct lustre_mdt_attrs *)info->oti_buf;
362 buf.lb_len = sizeof(info->oti_buf);
364 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
367 lustre_lma_swab(lma);
368 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
369 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
370 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
371 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
372 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
373 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
376 } else if (rc == -ENODATA) {
377 /* haven't initialize LMA xattr */
385 * Concurrency: no concurrent access is possible that early in object
388 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
389 const struct lu_object_conf *conf)
391 struct osd_object *obj = osd_obj(l);
392 struct osd_device *osd = osd_obj2dev(obj);
397 LASSERT(osd_invariant(obj));
399 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
400 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
401 l->lo_header->loh_attr |= LOHA_EXISTS;
405 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
408 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
410 LASSERT(obj->oo_dn == NULL);
411 rc = __osd_obj2dnode(env, osd->od_os, oid, &obj->oo_dn);
412 /* EEXIST will be returned if object is being deleted in ZFS */
418 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
419 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
423 rc = osd_object_init0(env, obj);
427 rc = osd_check_lma(env, obj);
430 } else if (rc == -ENOENT) {
433 LASSERT(osd_invariant(obj));
439 * Concurrency: no concurrent access is possible that late in object
442 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
444 struct osd_object *obj = osd_obj(l);
446 LASSERT(osd_invariant(obj));
448 dt_object_fini(&obj->oo_dt);
449 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
453 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
457 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
459 /* the object is supposed to be exclusively locked by
460 * the caller (osd_object_destroy()), while the transaction
461 * (oh) is per-thread and not shared */
462 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
463 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
470 /* Default to max data size covered by a level-1 indirect block */
471 static unsigned long osd_sync_destroy_max_size =
472 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
473 module_param(osd_sync_destroy_max_size, ulong, 0444);
474 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
477 osd_object_set_destroy_type(struct osd_object *obj)
480 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
481 * only once and use it consistently thereafter.
483 down_write(&obj->oo_guard);
484 if (obj->oo_destroy == OSD_DESTROY_NONE) {
485 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
486 obj->oo_destroy = OSD_DESTROY_SYNC;
487 else /* Larger objects are destroyed asynchronously */
488 obj->oo_destroy = OSD_DESTROY_ASYNC;
490 up_write(&obj->oo_guard);
493 static int osd_declare_object_destroy(const struct lu_env *env,
494 struct dt_object *dt,
497 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
498 struct osd_object *obj = osd_dt_obj(dt);
499 struct osd_device *osd = osd_obj2dev(obj);
500 struct osd_thandle *oh;
506 LASSERT(dt_object_exists(dt));
508 oh = container_of0(th, struct osd_thandle, ot_super);
509 LASSERT(oh->ot_tx != NULL);
511 /* declare that we'll remove object from fid-dnode mapping */
512 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
513 dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, NULL);
515 osd_declare_xattrs_destroy(env, obj, oh);
518 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
519 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
523 /* data to be truncated */
524 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
525 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
529 osd_object_set_destroy_type(obj);
530 if (obj->oo_destroy == OSD_DESTROY_SYNC)
531 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
534 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
536 /* will help to find FID->ino when this object is being
537 * added to PENDING/ */
538 osd_idc_find_and_init(env, osd, obj);
543 static int osd_object_destroy(const struct lu_env *env,
544 struct dt_object *dt, struct thandle *th)
546 struct osd_thread_info *info = osd_oti_get(env);
547 char *buf = info->oti_str;
548 struct osd_object *obj = osd_dt_obj(dt);
549 struct osd_device *osd = osd_obj2dev(obj);
550 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
551 struct osd_thandle *oh;
556 down_write(&obj->oo_guard);
558 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
559 GOTO(out, rc = -ENOENT);
561 LASSERT(obj->oo_dn != NULL);
563 oh = container_of0(th, struct osd_thandle, ot_super);
565 LASSERT(oh->ot_tx != NULL);
567 /* remove obj ref from index dir (it depends) */
568 zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
569 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
571 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
572 osd->od_svname, buf, rc);
576 rc = osd_xattrs_destroy(env, obj, oh);
578 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
579 osd->od_svname, buf, rc);
583 oid = obj->oo_dn->dn_object;
584 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
585 /* this may happen if the destroy wasn't declared
586 * e.g. when the object is created and then destroyed
587 * in the same transaction - we don't need additional
588 * space for destroy specifically */
589 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
590 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
592 CERROR("%s: failed to free %s %llu: rc = %d\n",
593 osd->od_svname, buf, oid, rc);
594 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
595 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
597 CERROR("%s: failed to free %s %llu: rc = %d\n",
598 osd->od_svname, buf, oid, rc);
599 } else { /* asynchronous destroy */
600 rc = osd_object_unlinked_add(obj, oh);
604 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
607 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
608 osd->od_svname, buf, oid, rc);
612 /* not needed in the cache anymore */
613 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
615 obj->oo_destroyed = 1;
616 up_write(&obj->oo_guard);
620 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
622 struct osd_object *obj = osd_obj(l);
624 if (obj->oo_dn != NULL) {
625 osd_object_sa_fini(obj);
626 if (obj->oo_sa_xattr) {
627 nvlist_free(obj->oo_sa_xattr);
628 obj->oo_sa_xattr = NULL;
630 osd_dnode_rele(obj->oo_dn);
631 list_del(&obj->oo_sa_linkage);
637 * Concurrency: ->loo_object_release() is called under site spin-lock.
639 static void osd_object_release(const struct lu_env *env,
645 * Concurrency: shouldn't matter.
647 static int osd_object_print(const struct lu_env *env, void *cookie,
648 lu_printer_t p, const struct lu_object *l)
650 struct osd_object *o = osd_obj(l);
652 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
655 static void osd_object_read_lock(const struct lu_env *env,
656 struct dt_object *dt, unsigned role)
658 struct osd_object *obj = osd_dt_obj(dt);
660 LASSERT(osd_invariant(obj));
662 down_read_nested(&obj->oo_sem, role);
665 static void osd_object_write_lock(const struct lu_env *env,
666 struct dt_object *dt, unsigned role)
668 struct osd_object *obj = osd_dt_obj(dt);
670 LASSERT(osd_invariant(obj));
672 down_write_nested(&obj->oo_sem, role);
675 static void osd_object_read_unlock(const struct lu_env *env,
676 struct dt_object *dt)
678 struct osd_object *obj = osd_dt_obj(dt);
680 LASSERT(osd_invariant(obj));
681 up_read(&obj->oo_sem);
684 static void osd_object_write_unlock(const struct lu_env *env,
685 struct dt_object *dt)
687 struct osd_object *obj = osd_dt_obj(dt);
689 LASSERT(osd_invariant(obj));
690 up_write(&obj->oo_sem);
693 static int osd_object_write_locked(const struct lu_env *env,
694 struct dt_object *dt)
696 struct osd_object *obj = osd_dt_obj(dt);
699 LASSERT(osd_invariant(obj));
701 if (down_write_trylock(&obj->oo_sem)) {
703 up_write(&obj->oo_sem);
708 static int osd_attr_get(const struct lu_env *env,
709 struct dt_object *dt,
710 struct lu_attr *attr)
712 struct osd_object *obj = osd_dt_obj(dt);
717 down_read(&obj->oo_guard);
719 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
720 GOTO(out, rc = -ENOENT);
722 LASSERT(osd_invariant(obj));
725 read_lock(&obj->oo_attr_lock);
726 *attr = obj->oo_attr;
727 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
728 attr->la_flags |= LUSTRE_ORPHAN_FL;
729 read_unlock(&obj->oo_attr_lock);
731 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
732 * from within sa_object_size() can block on a mutex, so
733 * we can't call sa_object_size() holding rwlock */
734 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
735 /* we do not control size of indices, so always calculate
736 * it from number of blocks reported by DMU */
737 if (S_ISDIR(attr->la_mode))
738 attr->la_size = 512 * blocks;
739 /* Block size may be not set; suggest maximal I/O transfers. */
741 blksize = osd_spa_maxblocksize(
742 dmu_objset_spa(osd_obj2dev(obj)->od_os));
744 attr->la_blksize = blksize;
745 attr->la_blocks = blocks;
746 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
749 up_read(&obj->oo_guard);
753 /* Simple wrapper on top of qsd API which implement quota transfer for osd
754 * setattr needs. As a reminder, only the root user can change ownership of
755 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
756 static inline int qsd_transfer(const struct lu_env *env,
757 struct qsd_instance *qsd,
758 struct lquota_trans *trans, int qtype,
759 __u64 orig_id, __u64 new_id, __u64 bspace,
760 struct lquota_id_info *qi)
764 if (unlikely(qsd == NULL))
767 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
768 qi->lqi_type = qtype;
770 /* inode accounting */
771 qi->lqi_is_blk = false;
773 /* one more inode for the new owner ... */
774 qi->lqi_id.qid_uid = new_id;
776 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
777 if (rc == -EDQUOT || rc == -EINPROGRESS)
782 /* and one less inode for the current id */
783 qi->lqi_id.qid_uid = orig_id;;
785 /* can't get EDQUOT when reducing usage */
786 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
787 if (rc == -EINPROGRESS)
792 /* block accounting */
793 qi->lqi_is_blk = true;
795 /* more blocks for the new owner ... */
796 qi->lqi_id.qid_uid = new_id;
797 qi->lqi_space = bspace;
798 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
799 if (rc == -EDQUOT || rc == -EINPROGRESS)
804 /* and finally less blocks for the current owner */
805 qi->lqi_id.qid_uid = orig_id;
806 qi->lqi_space = -bspace;
807 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
808 /* can't get EDQUOT when reducing usage */
809 if (rc == -EINPROGRESS)
814 static int osd_declare_attr_set(const struct lu_env *env,
815 struct dt_object *dt,
816 const struct lu_attr *attr,
817 struct thandle *handle)
819 struct osd_thread_info *info = osd_oti_get(env);
820 struct osd_object *obj = osd_dt_obj(dt);
821 struct osd_device *osd = osd_obj2dev(obj);
823 struct osd_thandle *oh;
831 LASSERT(handle != NULL);
832 LASSERT(osd_invariant(obj));
834 oh = container_of0(handle, struct osd_thandle, ot_super);
836 down_read(&obj->oo_guard);
837 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
840 LASSERT(obj->oo_sa_hdl != NULL);
841 LASSERT(oh->ot_tx != NULL);
842 /* regular attributes are part of the bonus buffer */
843 /* let's check whether this object is already part of
846 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
847 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
848 if (txh->txh_dnode == NULL)
850 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
852 /* this object is part of the transaction already
853 * we don't need to declare bonus again */
858 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
859 if (oh->ot_tx->tx_err != 0)
860 GOTO(out, rc = -oh->ot_tx->tx_err);
862 if (attr && attr->la_valid & LA_FLAGS) {
863 /* LMA is usually a part of bonus, no need to declare
867 if (attr && (attr->la_valid & (LA_UID | LA_GID))) {
868 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
869 bspace = toqb(bspace * blksize);
872 if (attr && attr->la_valid & LA_UID) {
873 /* quota enforcement for user */
874 if (attr->la_uid != obj->oo_attr.la_uid) {
875 rc = qsd_transfer(env, osd->od_quota_slave,
876 &oh->ot_quota_trans, USRQUOTA,
877 obj->oo_attr.la_uid, attr->la_uid,
878 bspace, &info->oti_qi);
883 if (attr && attr->la_valid & LA_GID) {
884 /* quota enforcement for group */
885 if (attr->la_gid != obj->oo_attr.la_gid) {
886 rc = qsd_transfer(env, osd->od_quota_slave,
887 &oh->ot_quota_trans, GRPQUOTA,
888 obj->oo_attr.la_gid, attr->la_gid,
889 bspace, &info->oti_qi);
896 up_read(&obj->oo_guard);
901 * Set the attributes of an object
903 * The transaction passed to this routine must have
904 * dmu_tx_hold_bonus(tx, oid) called and then assigned
905 * to a transaction group.
907 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
908 const struct lu_attr *la, struct thandle *handle)
910 struct osd_thread_info *info = osd_oti_get(env);
911 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
912 struct osd_object *obj = osd_dt_obj(dt);
913 struct osd_device *osd = osd_obj2dev(obj);
914 struct osd_thandle *oh;
915 struct osa_attr *osa = &info->oti_osa;
916 __u64 valid = la->la_valid;
922 down_read(&obj->oo_guard);
923 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
924 GOTO(out, rc = -ENOENT);
926 LASSERT(handle != NULL);
927 LASSERT(osd_invariant(obj));
928 LASSERT(obj->oo_sa_hdl);
930 oh = container_of0(handle, struct osd_thandle, ot_super);
931 /* Assert that the transaction has been assigned to a
932 transaction group. */
933 LASSERT(oh->ot_tx->tx_txg != 0);
935 /* Only allow set size for regular file */
936 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
937 valid &= ~(LA_SIZE | LA_BLOCKS);
939 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
942 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
945 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
951 if (valid & LA_FLAGS) {
952 struct lustre_mdt_attrs *lma;
955 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
956 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
957 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
959 buf.lb_len = sizeof(info->oti_buf);
960 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
964 le32_to_cpu(lma->lma_incompat);
966 lustre_to_lma_flags(la->la_flags);
968 cpu_to_le32(lma->lma_incompat);
970 buf.lb_len = sizeof(*lma);
971 rc = osd_xattr_set_internal(env, obj, &buf,
977 CWARN("%s: failed to set LMA flags: rc = %d\n",
984 write_lock(&obj->oo_attr_lock);
986 if (valid & LA_ATIME) {
987 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
988 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
991 if (valid & LA_MTIME) {
992 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
993 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
996 if (valid & LA_CTIME) {
997 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
998 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1001 if (valid & LA_MODE) {
1002 /* mode is stored along with type, so read it first */
1003 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1004 (la->la_mode & ~S_IFMT);
1005 osa->mode = obj->oo_attr.la_mode;
1006 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1009 if (valid & LA_SIZE) {
1010 osa->size = obj->oo_attr.la_size = la->la_size;
1011 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1014 if (valid & LA_NLINK) {
1015 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1016 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1019 if (valid & LA_RDEV) {
1020 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1021 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1024 if (valid & LA_FLAGS) {
1025 osa->flags = attrs_fs2zfs(la->la_flags);
1026 /* many flags are not supported by zfs, so ensure a good cached
1028 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1029 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1032 if (valid & LA_UID) {
1033 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1034 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1037 if (valid & LA_GID) {
1038 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1039 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1042 obj->oo_attr.la_valid |= valid;
1043 write_unlock(&obj->oo_attr_lock);
1045 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1046 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1049 up_read(&obj->oo_guard);
1056 * XXX temporary solution.
1059 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1060 struct dt_object *parent, struct dt_object *child,
1065 ah->dah_parent = parent;
1066 ah->dah_mode = child_mode;
1068 if (parent != NULL && !dt_object_remote(parent)) {
1069 /* will help to find FID->ino at dt_insert("..") */
1070 struct osd_object *pobj = osd_dt_obj(parent);
1072 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1076 static int osd_declare_object_create(const struct lu_env *env,
1077 struct dt_object *dt,
1078 struct lu_attr *attr,
1079 struct dt_allocation_hint *hint,
1080 struct dt_object_format *dof,
1081 struct thandle *handle)
1083 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1084 struct osd_object *obj = osd_dt_obj(dt);
1085 struct osd_device *osd = osd_obj2dev(obj);
1086 struct osd_thandle *oh;
1093 switch (dof->dof_type) {
1097 if (obj->oo_dt.do_body_ops == NULL)
1098 obj->oo_dt.do_body_ops = &osd_body_ops;
1104 LASSERT(handle != NULL);
1105 oh = container_of0(handle, struct osd_thandle, ot_super);
1106 LASSERT(oh->ot_tx != NULL);
1108 /* this is the minimum set of EAs on every Lustre object */
1109 obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1110 sizeof(__u64) + /* VBR VERSION */
1111 sizeof(struct lustre_mdt_attrs); /* LMA */
1112 /* reserve 32 bytes for extra stuff like ACLs */
1113 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1115 switch (dof->dof_type) {
1117 dt->do_index_ops = &osd_dir_ops;
1119 /* for zap create */
1120 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1121 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1126 /* first, we'll create new object */
1127 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1135 /* and we'll add it to some mapping */
1136 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
1137 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, NULL);
1139 /* will help to find FID->ino mapping at dt_insert() */
1140 osd_idc_find_and_init(env, osd, obj);
1142 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1143 false, NULL, false);
1148 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1149 sa_handle_t *sa_hdl, dmu_tx_t *tx,
1150 struct lu_attr *la, uint64_t parent)
1152 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1153 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1162 gen = dmu_tx_get_txg(tx);
1164 ZFS_TIME_ENCODE(&now, crtime);
1166 osa->atime[0] = la->la_atime;
1167 osa->ctime[0] = la->la_ctime;
1168 osa->mtime[0] = la->la_mtime;
1169 osa->mode = la->la_mode;
1170 osa->uid = la->la_uid;
1171 osa->gid = la->la_gid;
1172 osa->rdev = la->la_rdev;
1173 osa->nlink = la->la_nlink;
1174 osa->flags = attrs_fs2zfs(la->la_flags);
1175 osa->size = la->la_size;
1178 * we need to create all SA below upon object create.
1180 * XXX The attribute order matters since the accounting callback relies
1181 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1182 * look up the UID/GID attributes. Moreover, the callback does not seem
1183 * to support the spill block.
1184 * We define attributes in the same order as SA_*_OFFSET in order to
1185 * work around the problem. See ORI-610.
1188 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1189 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1190 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1191 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1192 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1193 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1194 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1195 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1196 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1197 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1198 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1201 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1203 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1208 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1209 uint64_t oid, dnode_t **dnp)
1214 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1215 for (txh = list_tail(&tx->tx_holds); txh;
1216 txh = list_prev(&tx->tx_holds, txh)) {
1217 dnode_t *dn = txh->txh_dnode;
1222 if (dn->dn_object != oid)
1226 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1227 if (dn->dn_bonus == NULL)
1228 dbuf_create_bonus(dn);
1229 rw_exit(&dn->dn_struct_rwlock);
1233 LASSERT(dn->dn_handle);
1235 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1236 refcount_add(&dn->dn_holds, tag);
1237 atomic_inc_32(&dn->dn_dbufs_count);
1240 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1244 if (unlikely(*dnp == NULL))
1245 rc = __osd_obj2dnode(env, tx->tx_objset, oid, dnp);
1251 * The transaction passed to this routine must have
1252 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1253 * to a transaction group.
1255 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1256 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1258 struct osd_device *osd = osd_obj2dev(obj);
1259 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1260 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1263 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1264 * would get an additional ditto copy */
1265 if (unlikely(S_ISREG(la->la_mode) &&
1266 fid_seq_is_local_file(fid_seq(fid))))
1267 type = DMU_OTN_UINT8_METADATA;
1269 /* Create a new DMU object using the default dnode size. */
1270 oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1272 LASSERT(la->la_valid & LA_MODE);
1276 return osd_find_new_dnode(env, tx, oid, dnp);
1280 * The transaction passed to this routine must have
1281 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1282 * to a transaction group.
1284 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1285 * This is fine for directories today, because storing the FID in the dirent
1286 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1287 * then we might need to re-evaluate the use of this flag and instead do
1288 * a conversion from the different internal ZAP hash formats being used. */
1289 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1290 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1295 /* Assert that the transaction has been assigned to a
1296 transaction group. */
1297 LASSERT(tx->tx_txg != 0);
1300 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1301 DMU_OT_DIRECTORY_CONTENTS,
1302 14, /* == ZFS fzap_default_blockshift */
1303 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1309 return osd_find_new_dnode(env, tx, oid, dnp);
1312 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1313 struct lu_attr *la, struct osd_thandle *oh)
1318 /* Index file should be created as regular file in order not to confuse
1319 * ZPL which could interpret them as directory.
1320 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1322 LASSERT(S_ISREG(la->la_mode));
1323 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1324 ZAP_FLAG_UINT64_KEY);
1330 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1331 struct lu_attr *la, struct osd_thandle *oh)
1336 LASSERT(S_ISDIR(la->la_mode));
1337 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la, 0);
1343 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1344 struct lu_attr *la, struct osd_thandle *oh)
1346 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1347 struct osd_device *osd = osd_obj2dev(obj);
1351 LASSERT(S_ISREG(la->la_mode));
1352 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1356 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
1358 /* The minimum block size must be at least page size otherwise
1359 * it will break the assumption in tgt_thread_big_cache where
1360 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1361 * RDMA due to subpage transfer size */
1362 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1363 PAGE_SIZE, 0, oh->ot_tx);
1365 CERROR("%s: can't change blocksize: %d\n",
1366 osd->od_svname, rc);
1374 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1375 struct lu_attr *la, struct osd_thandle *oh)
1380 LASSERT(S_ISLNK(la->la_mode));
1381 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1387 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1388 struct lu_attr *la, struct osd_thandle *oh)
1393 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1394 la->la_valid |= LA_RDEV;
1396 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1402 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1403 struct osd_object *obj,
1405 struct osd_thandle *oh);
1407 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1409 osd_obj_type_f result;
1435 * Concurrency: @dt is write locked.
1437 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1438 struct lu_attr *attr,
1439 struct dt_allocation_hint *hint,
1440 struct dt_object_format *dof,
1443 struct osd_thread_info *info = osd_oti_get(env);
1444 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1445 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1446 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1447 struct osd_object *obj = osd_dt_obj(dt);
1448 struct osd_device *osd = osd_obj2dev(obj);
1449 char *buf = info->oti_str;
1450 struct osd_thandle *oh;
1452 uint64_t zapid, parent = 0;
1457 /* concurrent create declarations should not see
1458 * the object inconsistent (db, attr, etc).
1459 * in regular cases acquisition should be cheap */
1460 down_write(&obj->oo_guard);
1462 if (unlikely(dt_object_exists(dt)))
1463 GOTO(out, rc = -EEXIST);
1465 LASSERT(osd_invariant(obj));
1466 LASSERT(dof != NULL);
1468 LASSERT(th != NULL);
1469 oh = container_of0(th, struct osd_thandle, ot_super);
1471 LASSERT(obj->oo_dn == NULL);
1473 /* to follow ZFS on-disk format we need
1474 * to initialize parent dnode properly */
1475 if (hint != NULL && hint->dah_parent != NULL &&
1476 !dt_object_remote(hint->dah_parent))
1477 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1479 /* we may fix some attributes, better do not change the source */
1480 obj->oo_attr = *attr;
1481 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1483 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1491 zde->zde_dnode = dn->dn_object;
1492 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1494 zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
1496 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1500 /* Now add in all of the "SA" attributes */
1501 rc = osd_sa_handle_get(obj);
1505 /* configure new osd object */
1506 parent = parent != 0 ? parent : zapid;
1507 rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
1508 &obj->oo_attr, parent);
1512 /* XXX: oo_lma_flags */
1513 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1514 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1515 /* no body operations for accounting objects */
1516 obj->oo_dt.do_body_ops = &osd_body_ops;
1518 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1522 /* initialize LMA */
1523 lustre_lma_init(lma, lu_object_fid(&obj->oo_dt.do_lu), 0, 0);
1524 lustre_lma_swab(lma);
1525 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1526 (uchar_t *)lma, sizeof(*lma));
1529 rc = __osd_sa_xattr_update(env, obj, oh);
1532 osd_idc_find_and_init(env, osd, obj);
1535 if (unlikely(rc && dn)) {
1536 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1540 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1542 up_write(&obj->oo_guard);
1546 static int osd_declare_object_ref_add(const struct lu_env *env,
1547 struct dt_object *dt,
1550 return osd_declare_attr_set(env, dt, NULL, th);
1554 * Concurrency: @dt is write locked.
1556 static int osd_object_ref_add(const struct lu_env *env,
1557 struct dt_object *dt,
1558 struct thandle *handle)
1560 struct osd_object *obj = osd_dt_obj(dt);
1561 struct osd_thandle *oh;
1562 struct osd_device *osd = osd_obj2dev(obj);
1568 down_read(&obj->oo_guard);
1569 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1570 GOTO(out, rc = -ENOENT);
1572 LASSERT(osd_invariant(obj));
1573 LASSERT(obj->oo_sa_hdl != NULL);
1575 oh = container_of0(handle, struct osd_thandle, ot_super);
1577 write_lock(&obj->oo_attr_lock);
1578 nlink = ++obj->oo_attr.la_nlink;
1579 write_unlock(&obj->oo_attr_lock);
1581 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1584 up_read(&obj->oo_guard);
1588 static int osd_declare_object_ref_del(const struct lu_env *env,
1589 struct dt_object *dt,
1590 struct thandle *handle)
1592 return osd_declare_attr_set(env, dt, NULL, handle);
1596 * Concurrency: @dt is write locked.
1598 static int osd_object_ref_del(const struct lu_env *env,
1599 struct dt_object *dt,
1600 struct thandle *handle)
1602 struct osd_object *obj = osd_dt_obj(dt);
1603 struct osd_thandle *oh;
1604 struct osd_device *osd = osd_obj2dev(obj);
1610 down_read(&obj->oo_guard);
1612 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1613 GOTO(out, rc = -ENOENT);
1615 LASSERT(osd_invariant(obj));
1616 LASSERT(obj->oo_sa_hdl != NULL);
1618 oh = container_of0(handle, struct osd_thandle, ot_super);
1619 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1621 write_lock(&obj->oo_attr_lock);
1622 nlink = --obj->oo_attr.la_nlink;
1623 write_unlock(&obj->oo_attr_lock);
1625 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1628 up_read(&obj->oo_guard);
1632 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1633 __u64 start, __u64 end)
1635 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1638 /* XXX: no other option than syncing the whole filesystem until we
1639 * support ZIL. If the object tracked the txg that it was last
1640 * modified in, it could pass that txg here instead of "0". Maybe
1641 * the changes are already committed, so no wait is needed at all? */
1642 if (!osd->od_dt_dev.dd_rdonly)
1643 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1648 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1653 static struct dt_object_operations osd_obj_ops = {
1654 .do_read_lock = osd_object_read_lock,
1655 .do_write_lock = osd_object_write_lock,
1656 .do_read_unlock = osd_object_read_unlock,
1657 .do_write_unlock = osd_object_write_unlock,
1658 .do_write_locked = osd_object_write_locked,
1659 .do_attr_get = osd_attr_get,
1660 .do_declare_attr_set = osd_declare_attr_set,
1661 .do_attr_set = osd_attr_set,
1662 .do_ah_init = osd_ah_init,
1663 .do_declare_create = osd_declare_object_create,
1664 .do_create = osd_object_create,
1665 .do_declare_destroy = osd_declare_object_destroy,
1666 .do_destroy = osd_object_destroy,
1667 .do_index_try = osd_index_try,
1668 .do_declare_ref_add = osd_declare_object_ref_add,
1669 .do_ref_add = osd_object_ref_add,
1670 .do_declare_ref_del = osd_declare_object_ref_del,
1671 .do_ref_del = osd_object_ref_del,
1672 .do_xattr_get = osd_xattr_get,
1673 .do_declare_xattr_set = osd_declare_xattr_set,
1674 .do_xattr_set = osd_xattr_set,
1675 .do_declare_xattr_del = osd_declare_xattr_del,
1676 .do_xattr_del = osd_xattr_del,
1677 .do_xattr_list = osd_xattr_list,
1678 .do_object_sync = osd_object_sync,
1679 .do_invalidate = osd_invalidate,
1682 static struct lu_object_operations osd_lu_obj_ops = {
1683 .loo_object_init = osd_object_init,
1684 .loo_object_delete = osd_object_delete,
1685 .loo_object_release = osd_object_release,
1686 .loo_object_free = osd_object_free,
1687 .loo_object_print = osd_object_print,
1688 .loo_object_invariant = osd_object_invariant,
1691 static int osd_otable_it_attr_get(const struct lu_env *env,
1692 struct dt_object *dt,
1693 struct lu_attr *attr)
1699 static struct dt_object_operations osd_obj_otable_it_ops = {
1700 .do_attr_get = osd_otable_it_attr_get,
1701 .do_index_try = osd_index_try,